def run()

in senteval/trec.py [0:0]

• 36 lines of code
• 7 McCabe index (conditional complexity)

    def run(self, params, batcher):
        train_embeddings, test_embeddings = [], []

        # Sort to reduce padding
        sorted_corpus_train = sorted(zip(self.train['X'], self.train['y']),
                                     key=lambda z: (len(z[0]), z[1]))
        train_samples = [x for (x, y) in sorted_corpus_train]
        train_labels = [y for (x, y) in sorted_corpus_train]

        sorted_corpus_test = sorted(zip(self.test['X'], self.test['y']),
                                    key=lambda z: (len(z[0]), z[1]))
        test_samples = [x for (x, y) in sorted_corpus_test]
        test_labels = [y for (x, y) in sorted_corpus_test]

        # Get train embeddings
        for ii in range(0, len(train_labels), params.batch_size):
            batch = train_samples[ii:ii + params.batch_size]
            embeddings = batcher(params, batch)
            train_embeddings.append(embeddings)
        train_embeddings = np.vstack(train_embeddings)
        logging.info('Computed train embeddings')

        # Get test embeddings
        for ii in range(0, len(test_labels), params.batch_size):
            batch = test_samples[ii:ii + params.batch_size]
            embeddings = batcher(params, batch)
            test_embeddings.append(embeddings)
        test_embeddings = np.vstack(test_embeddings)
        logging.info('Computed test embeddings')

        config_classifier = {'nclasses': 6, 'seed': self.seed,
                             'usepytorch': params.usepytorch,
                             'classifier': params.classifier,
                             'kfold': params.kfold}
        clf = KFoldClassifier({'X': train_embeddings,
                               'y': np.array(train_labels)},
                              {'X': test_embeddings,
                               'y': np.array(test_labels)},
                              config_classifier)
        devacc, testacc, _ = clf.run()
        logging.debug('\nDev acc : {0} Test acc : {1} \
            for TREC\n'.format(devacc, testacc))
        return {'devacc': devacc, 'acc': testacc,
                'ndev': len(self.train['X']), 'ntest': len(self.test['X'])}